J Comput Aided Mol Des (2007) 21:145–153 DOI 10.1007/s10822-006-9090-y

ORIGINAL PAPER

3D-QSAR study of hallucinogenic phenylalkylamines by using CoMFA approach

Zhuoyong Zhang Æ Liying An Æ Wenxiang Hu Æ Yuhong Xiang

Received: 4 July 2006 / Accepted: 22 October 2006 / Published online: 4 January 2007 Springer Science+Business Media B.V. 2006

Abstract The three-dimensional quantitative struc- Keywords 3D-QSAR Amphetamines CoMFA ture–activity relationship (3D-QSAR) has been stud- Hallucinogen Phenylethylamines ied on 90 hallucinogenic phenylalkylamines by the comparative molecular field analysis (CoMFA). Two conformations were compared during the modeling. Introduction Conformation I referred to the amino group close to ring position 6 and conformation II related to the Hallucinogens are substances that provoke strong amino group trans to the phenyl ring. Satisfactory mental and psychic changes including disorientation, results were obtained by using both conformations. derealization and depersonalization, giving rise to a There were still differences between the two models. variety of abnormal phenomena [1]. In some countries, The model based on conformation I got better statis- they are used as components of drugs. Some people tical results than the one about conformation II. And especially young people show a special interest and this may suggest that conformation I be preponderant may be addicted to these drugs for stimulation and self- when the hallucinogenic phenylalkylamines interact realization effects. And more seriously, hallucinogens with the receptor. To further confirm the predictive may be used to produce terror events by terrorists. To capability of the CoMFA model, 18 compounds with ensure the safety and the peace, strict administration conformation I were randomly selected as a test set and fast monitoring of hallucinogens are required. and the remaining ones as training set. The best Hallucinogens include mainly two categories of CoMFA model based on the training set had a cross- compounds according to their chemical structure. validation coefficient q2 of 0.549 at five components One is phenylalkylamines (phenylethylamines and and non cross-validation coefficient R2 of 0.835, the amphetamines), and the other is indolealkylamines standard error of estimation was 0.219. The model such as tryptamines and lysergic acid diethylamide showed good predictive ability in the external test with (LSD) derivatives .The 5-HT2A receptor is consid- 2 a coefficient Rpre of 0.611. The CoMFA coefficient ered to act as the biological target of these com- contour maps suggested that both steric and electro- pounds [2], however, the three-dimensional structure static interactions play an important role. The contri- of 5-HT2A receptor is not available so far, so most butions from the steric and electrostatic fields were studies are based on homologous compounds. To 0.450 and 0.550, respectively. understand the activity of hallucinogens at molecular level, their quantitative structure–activity relation- ships (QSARs) have been studied and several QSAR models have been established [3–7]. Most of the models were based on quantum chemistry parameters & Z. Zhang ( ) L. An W. Hu Y. Xiang or physical chemistry parameters, and the computa- Department of Chemistry, Capital Normal University, 105 Xisanhuan North, Beijing 100037, P.R. China tion based on quantum chemistry is slow and time- e-mail: [email protected] consuming.

123 146 J Comput Aided Mol Des (2007) 21:145–153

Comparative Molecular Field Analysis (CoMFA) USA) on a personal computer with Pentium IV pro- was developed by Cramer [8], and this method has cessor. Molecular building was done with molecular been well established for ligand-based 3D-QSAR sketch program. Since the crystal structure of DOET studies [9, 10]. It is based upon the calculated energies (4-ethyl-2,5-dimethoxyamphetamine) has been re- of steric and electrostatic interactions between the ported [11], the rest of the molecules were constructed compound and the probe atom placed at the various using DOET as template. Active conformation selec- intersections of a regular 3-D lattice. After this region tion is a key step for CoMFA analysis. Conformation is generated, the results are compared to the pharma- with the lowest energy is not always the active confor- cological data, and a linear combination of these two mation, and the proper active conformation can only be sets of data is constructed using a partial least squares extracted from the crystal structure of the complex of (PLS) algorithm. Cross-validated and non-cross- the drug molecule and its binding receptor [12–13]. validated r2-values are determined based on the PLS There are mainly two lower conformations for these results in order to validate the predictive properties of compounds (Fig. 1), conformation I referred to the the model. The r2-values can be optimized by itera- amino group close to ring position 6 and conformation tively varying the alignment rules, conformations and II related to the amino group trans to the phenyl ring. other parameters inherent to the technique. Energy comparison showed that conformation I was In this paper, the 3D-QSAR models of 90 phen- more stable than conformation II. To better understand ylalkylamines were established by CoMFA method. the activities of these compounds, the both conforma- This general procedure has been used in the present tions were used to build the models. The crystal con- study to gain insight into the steric and the electrostatic formation of DOET is similar to conformation I. The properties of these hallucinogenic phenylalkylamines, second conformation was acquired by interchange the their influence on the activity and to derive predictive positions of amino group and the methyl group 3D-QSAR models for discovery and prediction of the (amphetamines) or hydrogen atom (phenylethylam- hallucinogenic activities of new analogs for this class of ines). Partial atomic charges were assigned to each hallucinogens. atom and then energy minimization of each molecule was performed using Powell method and Tripos stan- dard force field with a distance-dependent dielectric Data set and methodology function. The minimization was terminated when the energy gradient convergence criterion of 0.005 kcal/ Biological data mol was reached or when the 2000-step minimization cycle limit was exceeded. The structures of 90 phenylalkylamine compounds and Molecular alignment is considered as one of the the biological activities data were cited from the ref- most sensitive parameters in CoMFA analysis [14]. erence [5]. This biological data were collected by The quality and the predictive ability of the model are Shulgin and co-workers. The reported work was about directly dependent on the alignment rule. Once the the hallucinogenic effect on human (oral activity data) active conformation was determined, pharmacophore and it has been considered as a benchmark in QSAR or common substructure alignment was carried out studies. The activities (in Mescaline units) were the according to some rules. In this paper, common sub- ratio of the effective dose of mescaline (350 mg) to the structure alignment was carried out using database mean of the threshold dose of the trial drug and the alignment tool with compound DOB as the template dose required to obtain the full effect. In the following molecule (Fig. 2). To refine the superimposition, some discussion, the alphabet A stands for the relative bio- molecules were manually adjusted using the rotation logical activity, Logarithm of A (log A) was applied in tools. Alignment of all compounds was shown in Fig. 3. the process of modeling. The structures and activity data of all the compounds were sorted by their struc- CoMFA analysis ture character and were listed in Tables 1a, b and c, respectively. Eighteen compounds were randomly In 3D-QSAR analysis, all aligned molecules were put selected as test set marked with asterisks in the tables. into a 3D cubic lattice that extending at least 0.4 nm beyond the volumes of all investigated molecules on all Molecular structure building and alignment axes. The region was partitioned into hundreds of grids with certain grid spacing. In the CoMFA analysis, All computational studies were performed by the Lennard-Jones 6–12 and Coulomb potentials were molecular modeling package sybyl 7.1 (Tripos Inc., employed to calculate the CoMFA steric and

123 J Comput Aided Mol Des (2007) 21:145–153 147

Table 1 Structure and hallucinogenic activity of (a) Phenylethylamines, (b) amphetamines and (c) some specially substituted compounds No Designation R A Log A

(a) Phenylethylamines [5] 1 Mescaline 3,4,5-trimethoxy 1 0.00 2ME* 3-ethoxy-4, 5-dimethoxy 1 0.00 3 E 3,5-dimethoxy-4-ethoxy 6 0.78 4 P 3,5-dimethoxy-4-propoxy 7 0.85 5 ASB 3,4-diethoxy-5-methoxy 1.3 0.11 6 2C-E 2,5-dimethoxy-4-ethyl 18 1.26 7 2C-D 2,5-dimethoxy-4-methyl 8 0.90 8 3-TM 3,4-dimethoxy-5-methylthio 4 0.60 9 TM 3,5-dimethoxy-4-methylthio 10 1.00 10 3-TME 3,4-dimethoxy-5-ethylthio 4 0.60 11 4-TME 3-ethoxy-4-methylthio-5-methoxy 4 0.60 12 3-TE* 3-methoxy-4-ethoxy-5-methylthio 4 0.60 13 4-TE 3,5-dimethoxy-4-ethylthio 12 1.08 14 3-TASB 3-ethylthio-4-ethoxy-5-methoxy 2 0.30 15 4-TASB 3-ethoxy-4-ethylthio-5-methoxy 4 0.60 16 5-TASB* 3,4-diethoxy-5-methylthio 2 0.30 17 TP 3,5-dimethoxy-4-propylthio 16 1.20 18 TB 3,5-dimethoxy-4-butylthio 3 0.48 19 2C-G 2,5-dimethoxy-3,4-dimethyl 11 1.04 20 2C-T-F* 2,5-dimethoxy-4-(2-fluoroethylthio) 29 1.46 21 2C-T-13 2,5-dimethoxy-4-(2-methoxyethylthio) 9 0.95 22 2C-B* 2,5-dimethoxy-4-bromo 16 1.20 23 2C-I 2,5-dimethoxy-4-iodine 17 1.23 24 2C-C 2,5-dimethoxy-4-chloro 10 1.00 25 CPM 3,5-dimethoxy-4-cyclopropylmethoxy 4 0.60 26 IP 3,5-dimethoxy-4-(i)-propoxy 5 0.70 27 BOD* 4-methyl-2,5, b-trimethoxy 15 1.18 28 BOH b-methoxy-3,4-methylenedioxy 3 0.48 29 2C-P 2,5-dimethoxy-4-n-propyl 37 1.57 30 2C-T* 2,5-dimethoxy-4-methylthio 4 0.60 31 2C-T-2 2,5-dimethoxy-4-ethylthio 16 1.20 32 2C-T-4 2,5-dimethoxy-4-isopropylthio 21 1.32 33 2C-T-7 2,5-dimethoxy-4-n-propylthio 15 1.18 34 2C-T-8 2,5-dimethoxy-4-cyclopropylmethylthio 7 0.85 35 2C-T-9 2,5-dimethoxy-4-tert-butylthio 4 0.60 36 2C-T-17 2,5-dimethoxy-4-sec-butylthio 4 0.60 37 HOT-2* 2,5-dimethoxy-4-ethylthio-N-hydroxy 22 1.34 38 HOT-7 2,5-dimethoxy-N-hydroxy-4-n-propylthio 15 1.18 39 HOT-17 2,5-dimethoxy-4-sec-butylthio-N-hydroxy 3 0.48 40 BOB 2,5,b-trimethoxy-4-bromo 20 1.30 41 DESOXY 3,5-dimethoxy-4-methyl 4 0.60 42 MAL 3,5-dimethoxy-4-methallyloxy 6 0.78 43 MDPH a, a-dimethyl-3, 4-methylenedioxy 1.5 0.18 (b) Amphetamines 44 4-MA* 4-methoxy 5 0.70 45 2,5-DMA 2,5-dimethoxy 2.5 0.40 46 TMA 3,4,5-trimethoxy 2 0.30 47 TMA-2 2,4,5-trimethoxy 10 1.00 48 TMA-4 2,3,5-trimethoxy 4 0.60 49 TMA-5 2,3,6-trimethoxy 10 1.00 50 TMA-6* 2,4,6-trimethoxy 10 1.00 51 MEM* 2,5-dimethoxy-4-ethoxy 9 0.95 52 3C-BZ 3,5-dimethoxy-4-benzyloxy 3 0.48 53 TA 2,3,4,5-tetramethoxy 6 0.78 54 MDA* 3,4-methylenedioxy 3 0.48 55 MMDA 3-methoxy-4, 5-methylenedioxy 2 0.30 56 MMDA-2 2-methoxy-4, 5-methylenedioxy 8 0.90 57 MMDA-3A 2-methoxy-3, 4-methylenedioxy 6 0.78 58 DMMDA 2,5-dimethoxy-3, 4-methylenedioxy 6 0.78 59 DOM 2,5-dimethoxy-4-methyl 50 1.70

123 148 J Comput Aided Mol Des (2007) 21:145–153

Table 1 continued No Designation R A Log A

60 DOET 2,5-dimethoxy-4-ethyl 75 1.88 61 DOEF 2,5-dimethoxy-4-(2-fluoroethyl) 110 2.04 62 DOPR* 2,5-dimethoxy-4- (n)-propyl 80 1.90 63 PARADOT 2,5-dimethoxy-4-methylthio 40 1.60 64 ALEPH-4 2,5-dimethoxy-4-(i)-propylthio 32 1.51 65 DOB 2,5-dimethoxy-4-bromo 150 2.18 66 DOI* 2,5-dimethoxy-4-iodine 133 2.12 67 DOC 2,5-dimethoxy-4-chloro 133 2.12 68 DON 2,5-dimethoxy-4-nitro 80 1.90 69 AL 3,5-dimethoxy-4-allyloxy 11 1.04 70 ALEPH-2* 2,5-dimethoxy-4-ethylthio 50 1.70 71 Y-DOM 2,6-dimethoxy-4-methyl 15 1.18 72 4-Br-3,5-DMA* 3,5-dimethoxy-4-bromo 43 1.63 73 FLEA N-hydroxy-N-methyl-3,4-dimethylenedioxy 2.5 0.40 74 G-3 2,5-dimethoxy-3,4-trimethylene 20 1.30 75 G* 2,5-dimethoxy-3,4-dimethyl 12 1.08 76 MDE 3,4-methylenedioxy-N-ethyl 2 0.30 77 MDOH 3,4-methylenedioxy-N-hydroxy 2.3 0.36 78 5-TOET 4-ethyl-2-methoxy-5-methylthio 15 1.18 79 2-TOM 5-methoxy-4-methyl-2-methylthio 4 0.60 80 5-TOM 2-methoxy-4-methyl-5-methylthio 7 0.85 81 ALEPH-7 4-propylthio-2,5-dimethoxy 55 1.74 82 3C-E 3,5-dimethoxy-4-ethoxy 7 0.85 83 META-DOB 2,4-dimethoxy-5-bromo 4 0.60 (c) Some specially substituted compounds 84 2C-G-3 O 14 1.15

NH2

O

85 2C-G-5 O 22 1.34

NH2

O

86 2C-G-N O 10 1.00

NH2

O

87 DMCPA O 17 1.23

NH2

O

123 J Comput Aided Mol Des (2007) 21:145–153 149

Table 1 continued No Designation R A Log A

88 G-5 18 1.26 O

NH2

O

NH2 89 J* O 1.5 0.18

O

H N 90 Methyl-J O 1.5 0.18

O

R 2 R' R 3

R

R 4

R 5 R 6

’ Conformation I: R =NH2, R=CH3/H ’ Conformation II: R = CH3/H, R= NH2

Fig. 1 Structure of phenylalkylamines N Fig. 3 Alignment of compounds in the training set

mal number of components produces the smallest root Fig. 2 Basic substructure for alignment mean predictive sum of squared errors, which usually corresponds to the highestP cross-validated squared 2 2 2 P ðYobsYpreÞ 3 coefficient (q ) q ¼ 1 2. The predictive electrostatic interaction fields, respectively. An sp - ðYobsYmeanÞ hybrized carbon atom with a charge of +1 was used as power of the resulting 3D-QSAR models was assessed 2 the probe atom and the steric and electrostatic energy on the holdout test set using the Rpre metric defined as cutoff was 30 kcal/mol and column-filtering value set to 2 SDPRESS Rpre = SD , where SD is defined as the sum of 1.0 cal/mol. squared deviations between the biological activities of Partial least squares method was carried out with the the test set and the mean value of the training set leave-one-out (LOO) cross-validation procedure to responses, and PRESS is the sum of the squared determine the optimum number of components for the deviation between the predicted and experimental final non-cross-validated 3D-QSAR models. The opti- bioactivities for the test compounds.

123 150 J Comput Aided Mol Des (2007) 21:145–153

Results and discussion Table 3 The influence of different grid spacing Grid spacing q2 NR2 sF Conformation analysis 1.0 0.669 6 0.858 0.201 83.789 The statistical results of the two models from different 1.5 0.649 6 0.847 0.209 76.453 conformations were summarized in Table 2. It sug- 2 0.640 7 0.879 0.187 85.102 2.5 0.589 6 0.810 0.232 59.147 gested that both CoMFA models have predictive ability (q2 > 0.50). Contributions of the steric and Notations are the same as Table 2 electrostatic fields in the two models were almost the same. But the model based on conformation I was a expressed. In this paper, smaller step size increased the little better. This may suggest that conformation I q2 value but not very significantly. Moreover, smaller should be preponderant when the ligands interacted step size dramatically increased the computation and with the 5-HT2Areceptor. And this hypothesis was in much more time was involved. So grid spacing was set agreement with Nichols’ work. to 0.2 nm in the following modeling. Nichols et al. have developed a homology model of the 5-HT2A receptor based on an in silico activated Selection of charge calculation method form of bovine rhodopsin [15]. When mescaline was docked into this model, it was observed that the 3- and In this paper, four charge calculation methods includ- 5-methoxy groups adopted out-of-plane conformations ing Gasteiger–Marsili, Pullman, Gasteiger–Hu¨ ckle and allowing hydrogen bonding with serine residues, and MMFF94 were compared. The results in Table 4 the side chain was in an approximately gauche con- showed that Gasteiger–Hu¨ ckle and MMFF94 charge formation. Next, they synthesized a more conforma- got better results than the other two. The worst results tionally restricted compound, which maintains the were obtained with Gasteiger–Marsili charge method. embedded mescaline structure in a low-energy con- This case may be due to the Gasteiger–Marsili charge formation and docked this molecule to the 5-HT2A only accounts for r electron, but both r and p electrons homology. And it also showed that the amine-bearing contribute to the interactions between receptor and side chain adopts a gauche orientation upon binding to ligand molecules. MMFF94 charge brought a larger the receptor [16]. So, we can conclude that conforma- standard error than Gasteiger–Hu¨ ckle charge, so tion I is more favorable. Gasteiger–Hu¨ ckle charge was utilized.

Selection of grid spacing CoMFA results

The influence of different grid spacing was investigated The statistical results of the two models were summa- and the results were listed in Table 2. It can be con- rized in Table 2. It suggested that both CoMFA models cluded from Table 3 that grid spacing affected the had predictive ability (q2 > 0.50). But the model from model to certain extent. If grid spacing is too large, the conformation I got higher q2 value than the other one grids become sparse, which may cause some important by nearly 10%. Both of the models got good results at molecular field information being unable to be well seven principal components. And there is little differ- ence between the non cross-validation coefficients. But the model I got a much smaller standard error of Table 2 The statistic results of the two conformations estimation. Contributions of the steric and electrostatic Statistic index Conformation I Conformation II fields in the two models were almost the same. But the model based on conformation I was a little better. To N 77 q2 0.640 0.596 further study the predictive ability of this model, 18 R2 0.879 0.849 s 0.187 0.209 Table 4 The influence of different charge calculation methods F 85.102 65.914 Contributions % Charge q2 NR2 sF Steric 45.0 45.2 Electrostatic 55.0 54.8 Gasteiger–Hu¨ ckle 0.640 7 0.879 0.187 83.789 Gasteiger–Marsili 0.445 7 0.875 0.184 63.985 Note: q2 is the cross-validated squared coefficient, N is the 2 Pullman 0.489 6 0.815 0.222 47.843 optimal number of components, R is the non cross-validated MMFF94 0.605 6 0.840 0.209 59.175 squared coefficient, s is the standard error of estimation and F is the F-test value Notations are the same as Table 2

123 J Comput Aided Mol Des (2007) 21:145–153 151 compounds were randomly selected as the test set and and test sets. In Fig. 4, most points evenly distributed the remaining as the training set. The best model had a among the line Y = X, which suggested the good q2 of 0.549 at five components, non cross-validation quality of the models. squared coefficient of 0.835 and standard error of estimated value of 0.219. For the prediction of the CoMFA coefficient contour map analysis 2 holdout test compounds the Rpre was 0.611, these statistical results confirmed the predictive capacity of Figure 5 showed the steric and electrostatic fields of the resultant CoMFA model. The predictive values of compound ME (A = 1 MU, Mescaline Unit) and the holdout test compounds were listed in Table 5. The DOEF (A = 110 MU) based on the CoMFA model of best CoMFA model was represented in Fig. 4, which the 90 compounds. The green contours characterize the shows the correlation of experimental values versus the regions where bulky substituents would increase the predicted values for compounds both in the training biological activity, whereas yellow contours indicate regions where steric bulk would not be tolerated. The blue and red polyhedra depict the favorable sites for Table 5 Actual value and the corresponding predicted values for positively and negatively charged groups, respectively. the test compounds (activities are in Mescaline unit) The green polyhedron located at the 4-position of the Compounds Actual Predicted Residue phenyl ring indicates that bulky substituents would be favorable. This can explain compound E (only ethoxy 2C-B 1.204 1.199 0.005 substitutes of methoxy in molecule Mescaline at the 2C-T 0.602 1.062 –0.460 2C-T-F 1.462 1.748 –0.286 4-position) is five times higher than Mescaline. At the 3-TE 0.602 0.482 0.120 3-position, there is a relatively large yellow region, so 4-Br-3, 5-DMA 1.634 1.073 0.561 4-MA 0.699 0.667 0.032 5-TASB 0.301 0.235 0.066 ALEPH-2 1.699 1.622 0.077 BOD 1.176 1.064 0.112 DOI 2.124 1.846 0.278 DOPR 1.903 1.819 0.084 G 1.08 1.024 0.056 HOT-2 1.34 1.028 0.312 J 0.179 0.254 –0.075 MDA 0.477 0.255 0.222 ME 0 0.237 –0.237 MEM 0.954 1.303 –0.349 TMA-6 1 1.627 –0.627

2.0 test set training set

1.5

1.0 predicted value 0.5

0.0

0.0 0.5 1.0 1.5 2.0 actual value Fig. 5 Contour maps of the steric and electrostatic field of Fig. 4 Predicted versus actual value of compounds both in molecule ME (with low activity) and DOEF (with high activity) training and test set for the best CoMFA model based on the whole CoMFA model

123 152 J Comput Aided Mol Des (2007) 21:145–153

ME with an ethoxy group at 3-position has activity the O O same as Mescaline. The large blue polyhedra encircling NH NH the benzene ring at the meta-positions indicate nega- 2 2 tively charged elements for meta-substitution on phe- nyl ring were unfavorable, which appears to account Br Br for the poor activity of some compounds with methy- lenedioxy-substituents such as MDE, MDOH, J, and O O BOH, etc. The red polyhedron located at the 4-posi- tion of the phenyl ring suggests that negative atom or compound 1 compound 2 group may increase activity, so, the electron-rich Fig. 6 Structures of two newly synthesized potential hallucino- groups and atoms at this position (such as DOEF, gens DOB, DOC, and DON) show strong hallucinogenic activity. Figure 5 distinctly showed the F atom located respectively. This similarity further conformed the in the red polyhedron. However, predictive values for predictive ability of the model. 4-halide substituted compounds such as DOB and DOC were a little lower than the actual values. These Conclusions results showed that there might exist other factors influencing the overall activities. And there is evidence In this study, CoMFA model has been created to that substituent at the 4-position of various phenyl- explain the observed structure–activity relationship for isopropylamines might directly interact with receptors a series of hallucinogenic phenylalkylamines. Two [17]; Hydrophobicity effect of the 4-position is crucial conformations were compared and the best model was to the hallucinogenic activity. Domelsmith et al. have achieved by the lower conformation (conformation I). examined a series of 13 phenylisopropylamine deriva- Several parameters were discussed in the optimization 2 tives, which revealed a correlation (r = 0.81) between of the model. The results of the holdout test showed their hallucinogenic potencies and the lipophilic values that this model had predictive ability. The coefficient of their 4-position substituents [18]. contour map showed that both steric and electrostatic There is another noticeable phenomenon that the fields play an important role, their contributions were molecule TM with methylthio substituent at the 4-po- 45.0% and 55.0%, respectively. Rational explanations sition of the phenyl ring has much higher activity than of the structure–activity relationship of these com- Mescaline (A = 10 MU). So it can be proposed that pounds were reached from the contour map. This alkylthio-substituent at the 4-position is favorable for offered important information for further structure the hallucinogenic activity. This ‘‘thio-effect’’ may modification and discovery new potential hallucino- result from the change of the orbital hybridization of genic phenylalkylamines. the heteroatom and therefore a change of the avail- ability of the molecule to metabolism [19]. Acknowledgements This work is supported by the Science and From the contour map, it can be found that sub- Technology Program of Beijing Municipal Government and the Scientific Research Common Program of Commission of Edu- stituents at the para- and meta- positions were crucial cation. to the hallucinogenic activity of these compounds. So modifications can be focused on these positions. And also we can presume that there may be other factors References that affect the hallucinogenic activity. To validate the established model, two separately 1. Nichols DE (2004) Pharmacol Ther 101:131 synthesized potential hallucinogenic compounds [20] 2. Krebs-Thomson K, Paulus MP, Geyer MA (1998) Neuro- psychopharmacology 18:339 (Figure 6) were used for prediction. The predicted 3. Nichols DE (1981) J Pharm Sci 70:839 values suggest that compound 2 have higher activity 4. Gupta SP, Singh P, Bindal MC (1983) Chem Rev 83:633 than compound 1. This result is consistent with the 5. Clare BW (1990) J Med Chem 33:687 conclusion from reference [20]. It was reported in the 6. Clare BW (2002) Computer-Aided Mol Des 16:611 7. Schulze-Alexandru M, Kovar K-A (1999) Quant Struct-Act reference [20] that compound 2 has an approximately Relat 18:548 60-fold higher affinity than the compound 1 in vitro 8. Cramer RD, Patterson DE, Bunce JD (1988) J Am Chem assays. In the rat drug discrimination assay, the com- Soc 110:5959 pound 2 is three times more potent than the compound 9. Klebe G, Abraham U, Mietzner T (1994) J Med Chem 37:4130 1.The predicted hallucinogenic activity data of the two 10. Song MH, Breneman CM, Sukumar N (2004) Bioorg Med compounds from our model is 0.446 and 1.347, Chem 12:489

123 J Comput Aided Mol Des (2007) 21:145–153 153

11. Kennard O, Giacovazzo C, Horn AS, Mongiorgi R, Riva di 17. Glennon RA, Young R, Benington F, Morint RD (1982) J Sanseverino L (1974) J Chem Soc Perkin Trans II 10:1160 Med Chem 25:1163 12. Hu WX, Yun LH (1992) Chin Chem Lett 3:271 18. Domelsmith LN, Eaton TA, Houk KN, Anderson GM, 13. Pauling P, Data N (1980) J Pro Natl Acad Sci USA 77:708 Glennon RA, Shulgin AT, Castagnoli N, Kollman PA (1981) 14. Tripos Inc. (2005) Sybyl 7.1. Manual, St. Louis, MO J Med Chem 24:1414 15. Chambers JJ, Nichols DE (2002) J Comput-Aided Mol Des 19. Braun U, Braun G, Jacob P III, Nichols DE, Shulgin AT 16:511 (1978) NIDA Res Monogr 22:35 16. McLean TH, Chambers JJ, Parrish JC, Braden MR, Marona- 20. Parker MA, Lewicka DM, Lucaites VL, Nelson DL, Nichols Lewicka D, Kurrasch-Orbaugh D, Nichols DE (2006) J Med DE (1998) J Med Chem 41:5148 Chem 49:4269

123